User Archetype Discovery By Cluster Analysis of Caller Log Data: Tenure Evolution is Stable as Time Period Reduces

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Clustering analysis, or clustering, is an activity which can be applied to user event log data to determine the types of users which exist within a service, and can be used to gain insights into the client base by their behaviour. However, when applied to longitudinal user event log data, clustering can potentially misclassify regular users as ’one-off’ if their last interaction within their tenure of the service appears at the beginning of the observable data set. The main objective of this study was to investigate whether any impact of user tenure within longitudinal data on k-means clustering accuracy would occur. The current paper subjected a large telephony call log data set from a helpline to a k-means clustering algorithm to determine the types of callers that contact the helpline based on their usage characteristics (number of calls, mean duration of calls and variability of call duration). A threshold of one-month increments were applied to the data (callers appearing before the threshold but not after were removed each time) and then subsequently subjected to k-means clustering. Results showed that cluster structures remained stable after each threshold condition. Significant differences in cluster centers were found in one cluster across tenure conditions.
LanguageEnglish
Title of host publicationECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics
Subtitle of host publication''Design for Cognition''
EditorsMaurice Mulvenna, Raymond Bond
Place of PublicationNew York, NY, USA
Pages196-202
Number of pages7
ISBN (Electronic)9781450371667
DOIs
Publication statusPublished - 10 Sep 2019

Publication series

NameICPS
PublisherACM

Fingerprint

Cluster analysis
Clustering algorithms

Keywords

  • User Event Log Data
  • Clustering Analysis
  • K-Means Clustering
  • Telephony Call Log Data
  • Tenure
  • Helpline
  • Mental Health

Cite this

Turkington, R., Bond, RR., Mulvenna, M., Potts, C., O'Neill, S., & Armour, C. (2019). User Archetype Discovery By Cluster Analysis of Caller Log Data: Tenure Evolution is Stable as Time Period Reduces. In M. Mulvenna, & R. Bond (Eds.), ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics: ''Design for Cognition'' (pp. 196-202). (ICPS). New York, NY, USA. https://doi.org/10.1145/3335082.3335090
Turkington, Robin ; Bond, RR ; Mulvenna, Maurice ; Potts, Courtney ; O'Neill, Siobhan ; Armour, Cherie. / User Archetype Discovery By Cluster Analysis of Caller Log Data : Tenure Evolution is Stable as Time Period Reduces. ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics: ''Design for Cognition''. editor / Maurice Mulvenna ; Raymond Bond. New York, NY, USA, 2019. pp. 196-202 (ICPS).
@inproceedings{90a14db9e1454afb9bdf30c3a38d16e5,
title = "User Archetype Discovery By Cluster Analysis of Caller Log Data: Tenure Evolution is Stable as Time Period Reduces",
abstract = "Clustering analysis, or clustering, is an activity which can be applied to user event log data to determine the types of users which exist within a service, and can be used to gain insights into the client base by their behaviour. However, when applied to longitudinal user event log data, clustering can potentially misclassify regular users as ’one-off’ if their last interaction within their tenure of the service appears at the beginning of the observable data set. The main objective of this study was to investigate whether any impact of user tenure within longitudinal data on k-means clustering accuracy would occur. The current paper subjected a large telephony call log data set from a helpline to a k-means clustering algorithm to determine the types of callers that contact the helpline based on their usage characteristics (number of calls, mean duration of calls and variability of call duration). A threshold of one-month increments were applied to the data (callers appearing before the threshold but not after were removed each time) and then subsequently subjected to k-means clustering. Results showed that cluster structures remained stable after each threshold condition. Significant differences in cluster centers were found in one cluster across tenure conditions.",
keywords = "User Event Log Data, Clustering Analysis, K-Means Clustering, Telephony Call Log Data, Tenure, Helpline, Mental Health",
author = "Robin Turkington and RR Bond and Maurice Mulvenna and Courtney Potts and Siobhan O'Neill and Cherie Armour",
year = "2019",
month = "9",
day = "10",
doi = "10.1145/3335082.3335090",
language = "English",
isbn = "978-1-4503-7166-7",
series = "ICPS",
publisher = "ACM",
pages = "196--202",
editor = "Maurice Mulvenna and Raymond Bond",
booktitle = "ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics",

}

Turkington, R, Bond, RR, Mulvenna, M, Potts, C, O'Neill, S & Armour, C 2019, User Archetype Discovery By Cluster Analysis of Caller Log Data: Tenure Evolution is Stable as Time Period Reduces. in M Mulvenna & R Bond (eds), ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics: ''Design for Cognition''. ICPS, New York, NY, USA, pp. 196-202. https://doi.org/10.1145/3335082.3335090

User Archetype Discovery By Cluster Analysis of Caller Log Data : Tenure Evolution is Stable as Time Period Reduces. / Turkington, Robin; Bond, RR; Mulvenna, Maurice; Potts, Courtney; O'Neill, Siobhan; Armour, Cherie.

ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics: ''Design for Cognition''. ed. / Maurice Mulvenna; Raymond Bond. New York, NY, USA, 2019. p. 196-202 (ICPS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - User Archetype Discovery By Cluster Analysis of Caller Log Data

T2 - Tenure Evolution is Stable as Time Period Reduces

AU - Turkington, Robin

AU - Bond, RR

AU - Mulvenna, Maurice

AU - Potts, Courtney

AU - O'Neill, Siobhan

AU - Armour, Cherie

PY - 2019/9/10

Y1 - 2019/9/10

N2 - Clustering analysis, or clustering, is an activity which can be applied to user event log data to determine the types of users which exist within a service, and can be used to gain insights into the client base by their behaviour. However, when applied to longitudinal user event log data, clustering can potentially misclassify regular users as ’one-off’ if their last interaction within their tenure of the service appears at the beginning of the observable data set. The main objective of this study was to investigate whether any impact of user tenure within longitudinal data on k-means clustering accuracy would occur. The current paper subjected a large telephony call log data set from a helpline to a k-means clustering algorithm to determine the types of callers that contact the helpline based on their usage characteristics (number of calls, mean duration of calls and variability of call duration). A threshold of one-month increments were applied to the data (callers appearing before the threshold but not after were removed each time) and then subsequently subjected to k-means clustering. Results showed that cluster structures remained stable after each threshold condition. Significant differences in cluster centers were found in one cluster across tenure conditions.

AB - Clustering analysis, or clustering, is an activity which can be applied to user event log data to determine the types of users which exist within a service, and can be used to gain insights into the client base by their behaviour. However, when applied to longitudinal user event log data, clustering can potentially misclassify regular users as ’one-off’ if their last interaction within their tenure of the service appears at the beginning of the observable data set. The main objective of this study was to investigate whether any impact of user tenure within longitudinal data on k-means clustering accuracy would occur. The current paper subjected a large telephony call log data set from a helpline to a k-means clustering algorithm to determine the types of callers that contact the helpline based on their usage characteristics (number of calls, mean duration of calls and variability of call duration). A threshold of one-month increments were applied to the data (callers appearing before the threshold but not after were removed each time) and then subsequently subjected to k-means clustering. Results showed that cluster structures remained stable after each threshold condition. Significant differences in cluster centers were found in one cluster across tenure conditions.

KW - User Event Log Data

KW - Clustering Analysis

KW - K-Means Clustering

KW - Telephony Call Log Data

KW - Tenure

KW - Helpline

KW - Mental Health

UR - https://www.ulster.ac.uk/conference/european-conference-on-cognitive-ergonomics

UR - http://www.scopus.com/inward/record.url?scp=85073151932&partnerID=8YFLogxK

U2 - 10.1145/3335082.3335090

DO - 10.1145/3335082.3335090

M3 - Conference contribution

SN - 978-1-4503-7166-7

T3 - ICPS

SP - 196

EP - 202

BT - ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics

A2 - Mulvenna, Maurice

A2 - Bond, Raymond

CY - New York, NY, USA

ER -

Turkington R, Bond RR, Mulvenna M, Potts C, O'Neill S, Armour C. User Archetype Discovery By Cluster Analysis of Caller Log Data: Tenure Evolution is Stable as Time Period Reduces. In Mulvenna M, Bond R, editors, ECCE 2019 - Proceedings of the 31st European Conference on Cognitive Ergonomics: ''Design for Cognition''. New York, NY, USA. 2019. p. 196-202. (ICPS). https://doi.org/10.1145/3335082.3335090